Multi-species microbial communities can outperform single species in producing pharmaceuticals and fighting infections. However, for a community to be useful, it must be ?robust? in the sense that it must retain member species and survive internal and external perturbations. Community robustness arises from interactions between community members and can thus change rapidly as community members evolve. To date, we understand very little about various forms of robustness and how they change during evolution. Theoretical work is often based on unrealistic assumptions, while empirical work is largely observational or correlational. To understand community robustness and how they might change as community members evolve, we have created a cooperative yeast community. It consists of two mutually-dependent yeast strains exchanging essential metabolites. The two strains are reproductively isolated, and can thus be regarded as two species. Due to mutual dependence, the two strains coexist over a long term and can thus be further engineered to carry out ?division of labor? in complex tasks such as degrading a mixture of waste products. However, such a community can still go extinct upon population reduction. Here, we will examine community robustness against two commonly-encountered external perturbations: extreme population reduction such as during the colonization of a new host, or gradual population reduction such as during periodic purge from the gut. We aim to understand these two forms of robustness so that we can manipulate them. We also want to understand how robustness might change as community members evolve and diversify. We have passaged multiple communities for over 150 generations. All communities became more robust in surviving severe population reductions. Strikingly, robustness against gradual population reduction increased in some communities, but it decreased in other communities. To understand community robustness and how they change during evolution, we have developed high-throughput assays to measure phenotypes of cells from the two strains. Mathematical models based on these measurements successfully predicted for example robustness against severe population reduction in the ancestral community. We will use these mathematical models to predict how we might effectively alter robustness. We will also predict which subset of evolved community members are important in altering community robustness. We will experimentally test model predictions. Model-experiment discrepancies will motivate us to uncover missing elements that are important to community robustness, such as evolved new interactions and rare evolved genotypes with extreme phenotypes. Our work will provide an experimental and mathematical approach to understanding communities harboring evolutionary complexity.